CN113837847B - Knowledge intensive service recommendation method based on heterogeneous multivariate relation fusion - Google Patents
Knowledge intensive service recommendation method based on heterogeneous multivariate relation fusion Download PDFInfo
- Publication number
- CN113837847B CN113837847B CN202111276405.4A CN202111276405A CN113837847B CN 113837847 B CN113837847 B CN 113837847B CN 202111276405 A CN202111276405 A CN 202111276405A CN 113837847 B CN113837847 B CN 113837847B
- Authority
- CN
- China
- Prior art keywords
- service
- employer
- knowledge
- heterogeneous
- services
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 53
- 230000004927 fusion Effects 0.000 title claims abstract description 19
- 239000013598 vector Substances 0.000 claims abstract description 59
- 230000003993 interaction Effects 0.000 claims description 27
- 230000006870 function Effects 0.000 claims description 5
- 238000005295 random walk Methods 0.000 claims description 4
- 238000012549 training Methods 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 3
- 230000009286 beneficial effect Effects 0.000 abstract description 3
- 238000005065 mining Methods 0.000 abstract description 3
- 230000001737 promoting effect Effects 0.000 abstract 1
- 230000008569 process Effects 0.000 description 7
- 238000005070 sampling Methods 0.000 description 6
- 238000013461 design Methods 0.000 description 5
- 239000011159 matrix material Substances 0.000 description 5
- 230000015572 biosynthetic process Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 4
- 238000011084 recovery Methods 0.000 description 4
- 238000003786 synthesis reaction Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 230000007115 recruitment Effects 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000002679 ablation Methods 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 229940047586 chemet Drugs 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000013100 final test Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012821 model calculation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000012797 qualification Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- ACTRVOBWPAIOHC-XIXRPRMCSA-N succimer Chemical compound OC(=O)[C@@H](S)[C@@H](S)C(O)=O ACTRVOBWPAIOHC-XIXRPRMCSA-N 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/906—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Business, Economics & Management (AREA)
- Strategic Management (AREA)
- Marketing (AREA)
- Economics (AREA)
- Tourism & Hospitality (AREA)
- Computational Linguistics (AREA)
- Finance (AREA)
- Artificial Intelligence (AREA)
- Accounting & Taxation (AREA)
- Databases & Information Systems (AREA)
- Data Mining & Analysis (AREA)
- Primary Health Care (AREA)
- Biophysics (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Development Economics (AREA)
- Probability & Statistics with Applications (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Human Resources & Organizations (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Medical Treatment And Welfare Office Work (AREA)
Abstract
The application discloses a knowledge intensive service recommendation method based on heterogeneous multivariate relation fusion, which comprises the following steps of: s1, constructing a heterogeneous information network and acquiring a heterogeneous multi-element relation in the heterogeneous information network; s2, merging the heterogeneous polynomials, and learning comprehensive vector representations of employers and services; and S3, acquiring a predicted value based on the comprehensive vector representation of the employer and the service, and recommending the service to each employer according to the predicted value. The application can well realize employer knowledge-intensive service requirement mining in the industrial field, can help non-professional users facing information overload find interesting knowledge-intensive services in a large number of specialized applications, is beneficial to increasing the viscosity of employers to platforms, promoting sales and improving the service quality and profit of the platforms.
Description
Technical Field
The application relates to the technical field of recommendation, in particular to a knowledge intensive service recommendation method based on heterogeneous multivariate relation fusion.
Background
The recommendation method is used for predicting and recommending the products or services of interest to the user by various online platforms, and has outstanding effects in improving the sales rate and the exposure rate of the products or services and relieving information overload. The recommendation method of platforms such as Taobao, amazon and the like utilizes the historical transaction, browsing, clicking, evaluation and other behavior data of the user to analyze the preference of the user, predict and recommend the service or product most likely to be purchased by the user. But the existing recommendation methods are mainly used in the consumer field to recommend standardized products or services to users. These standardized products or services are used by a large number of users at the same time. Unlike standardized products or services, knowledge-intensive services such as industrial design, industrial intellectual manufacturing, market research, animation design, software development, etc. are highly customized according to employer needs, and the specific content of each knowledge-intensive service purchased by each employer is not disclosed externally due to privacy protection. Therefore, the existing recommendation methods for the consumer domain are not highly accurate for recommending knowledge-intensive services in the industrial domain.
The rapid development of service markets and information and communication technologies provides a basis for the online transaction of knowledge-intensive services. Platforms such as zbj.com, upwork, and 99designs focus on online transactions for service knowledge intensive services. With the increase of the online transaction frequency and the transaction number of the knowledge-intensive service, some recommendation methods are applied to the knowledge-intensive service platform to relieve the information overload problem. However, the existing recommendation method mainly helps workers find tasks highly related to interests and capabilities of the workers in a large number of customized knowledge-intensive service requirements, or recommends knowledge resources (information or human resources) to users according to requirement information submitted by the users. Currently, knowledge-intensive service platforms lack recommendation methods to explore employer potential knowledge-intensive service needs and recommend services. To promote sales, improve platform quality of service and profits, knowledge-intensive service recommendation methods are being developed to explore the relationships that exist between employer knowledge-intensive services (or employer knowledge-intensive service demand preferences) and recommend services.
Disclosure of Invention
Aiming at the defects in the prior art, the application provides a knowledge intensive service recommendation method based on heterogeneous multi-element relation fusion, which comprises the following steps:
s1, constructing a heterogeneous information network and acquiring a heterogeneous multi-element relation in the heterogeneous information network;
s2, merging the heterogeneous polynomials, and learning comprehensive vector representations of employers and services;
and S3, acquiring a predicted value based on the comprehensive vector representation of the employer and the service, and recommending the service to each employer according to the predicted value.
Preferably, in step S1, the method for constructing a heterogeneous information network includes the steps of:
s11: constructing a network mode of the knowledge intensive service information network to illustrate object types and interaction relations thereof in the knowledge intensive service information network; wherein the object types include employers, services, knowledge, business, and workers, and the interaction relationships include employer-services, knowledge-services, business-services, and workers-services;
s12, acquiring original data, preprocessing the original data, and establishing an object relation; wherein the object relationships include employer-service relationships, business-service relationships, worker-service relationships, service-service knowledge similarity relationships;
and S13, linking all object relations to construct a heterogeneous information network of the knowledge intensive service.
Preferably, in step S1, the method for obtaining the heterogeneous polynary relationship in the heterogeneous information network specifically includes: and designing a meta-path, and acquiring heterogeneous multi-element relations in the heterogeneous information network according to the designed meta-path by adopting a random walk strategy.
Preferably, in step S2, the method of learning the comprehensive vector representation of employer and service by fusing the heterogeneous multi-relations comprises the steps of:
s21, merging heterogeneous multi-element relations acquired by each element path, and learning employer and service semantic specific vector representations;
s22, merging the heterogeneous multi-element relations from all element paths, and learning the comprehensive vector representation of employer and service.
Preferably, in step S21, the specific method for learning the employer and service semantic specific vector representation by fusing the heterogeneous polynomials acquired by each element path is as follows: and inputting the heterogeneous polynomials acquired by each element path into a meta 2vec model for learning to obtain employer and service semantic specific vector representations.
Preferably, in step S22, the method of learning employer vector representations includes the steps of:
s221: converting the employer semantic-specific vector representation by the multi-layer perceptron;
s222: calculating the importance of each meta-path to each employer based on the employer semantic-specific vector representation;
s223: normalizing the importance of each element path to each employer, and calculating the weight of each path;
s224: and carrying out weighted fusion based on the weight of each element path to obtain the comprehensive vector representation of the employer.
Preferably, in step S3, a recommendation model is constructed based on the comprehensive vector representation of the employer and the service, and a predicted value is obtained based on the recommendation model, wherein the recommendation model is specifically:
wherein ,representing predicted value, p u Representing employer complex vector representations, q s Representing a service synthesis vector representation, h being the prediction layer.
Preferably, optimizing the model by using the loss function is further included before obtaining the predicted value by recommending the model, and the predicted value is minimizedAnd the actual value R us Difference between them.
Preferably, in step S3, the method for recommending services to each employer according to the predicted value is as follows: and sorting according to the size of the predicted value to obtain a recommendation list, and recommending the first K services of the recommendation list to the employer.
The beneficial effects of the application are as follows: the knowledge intensive service recommendation method based on heterogeneous multivariate relation fusion can well realize employer knowledge intensive service requirement mining in the industrial field. A large number of specialized applications are gathered on the knowledge-intensive service platform, and the knowledge-intensive service recommendation method is provided for the recommendation system, so that non-professional users facing information overload can find knowledge-intensive services interested by themselves in the large number of specialized applications, the viscosity of employers to the platform is increased, sales are promoted, and the service quality and profit of the platform are improved.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below. Like elements or portions are generally identified by like reference numerals throughout the several figures. In the drawings, elements or portions thereof are not necessarily drawn to scale.
FIG. 1 is a flowchart of a knowledge intensive service recommendation method based on heterogeneous multivariate relation fusion in accordance with the present application;
FIG. 2 is a schematic diagram of a network model of a knowledge intensive services information network;
FIG. 3 is a schematic diagram of creating an object relationship;
FIG. 4 is a segment of a knowledge-intensive service heterogeneous information network;
FIG. 5 is a schematic diagram of a meta-path of a design;
FIG. 6 is a graph illustrating the performance of a recommendation model under different meta-path combinations;
FIG. 7 is a graph of the performance of the recommendation model at different degrees of similarity;
FIG. 8 is a diagram showing recommendation models at different locationsThe following properties.
Detailed Description
Embodiments of the technical scheme of the present application will be described in detail below with reference to the accompanying drawings. The following examples are only for more clearly illustrating the technical aspects of the present application, and thus are merely examples, and are not intended to limit the scope of the present application.
It is noted that unless otherwise indicated, technical or scientific terms used herein should be given the ordinary meaning as understood by one of ordinary skill in the art to which this application belongs.
As shown in fig. 1, the knowledge intensive service recommendation method based on heterogeneous multi-element relation fusion provided by the application comprises the following steps:
s1, constructing a heterogeneous information network and acquiring a heterogeneous multi-element relation in the heterogeneous information network;
in step S1, the method for constructing the heterogeneous information network includes the following steps:
s11: constructing a network mode of the knowledge intensive service information network to illustrate object types and interaction relations thereof in the knowledge intensive service information network; wherein the object types include employers, services, knowledge, business, and workers, and the interaction relationships include employer-services, knowledge-services, business-services, and workers-services;
the interaction relationship is defined as follows:
1) Interaction relationship definition 1: knowledge-service interaction definition
Interaction relationship definition 1.1: some knowledge-intensive services purchased by the same employer are likely to require similar knowledge to perform.
Interaction relationship definition 1.2: knowledge intensive services requiring similar knowledge to perform are similar.
2) Interaction relationship definition 2: business activity-service interaction relationship definition
Interaction relationship definition 2.1: some knowledge-intensive services purchased by the same employer are likely to belong to the same business.
Interaction relationship definition 2.2: knowledge-intensive services that belong to the same business have a high degree of relevance.
3) Interaction relationship definition 3: worker-service interaction relationship definition
Interaction relationship definition 3.1: an employer may purchase knowledge-intensive services from familiar knowledge workers.
Interaction relationship definition 3.2: knowledge intensive services provided by the same knowledge worker have a high correlation.
According to the definition above, there are 5 object types in the information network of knowledge intensive services: employers, services, knowledge, business and knowledge workers, and 4 relationship types: employer-service, knowledge-service, business-service, and worker-service.
However, information describing the knowledge required to perform a service is rich and heterogeneous, such as: professions, academia, work experiences, project experiences, job title, acquired qualification certificates, and the like. If all knowledge information for performing various knowledge-intensive services is presented in the knowledge network of the knowledge-intensive service, such knowledge-service multi-linking relationships would make the knowledge network of the knowledge-intensive service cumbersome and the computational complexity of the model would increase. Therefore, knowledge data is utilized to quantify knowledge similarity among knowledge intensive services, a knowledge-service multi-link relationship is converted into a service-service knowledge similarity single-link relationship, an information network of the knowledge intensive services is lightened, and model calculation complexity is reduced.
The network model of the knowledge intensive service is shown in fig. 2. As can be seen from fig. 2, the information network of the knowledge intensive service includes 4 object types and 4 interactive relationship types, so the information network of the knowledge intensive service is a heterogeneous information network.
S12, acquiring original data, preprocessing the original data, and establishing an object relation; wherein the object relationships include employer-service relationships, business-service relationships, worker-service relationships, service-service knowledge similarity relationships; the process is as shown in fig. 3, where raw data is obtained and preprocessed, and an object relationship is established based on the preprocessed raw data, where the raw data includes transaction data, worker-service relationships, service-category relationships, knowledge data, 1) employer-service relationships: transaction data is the most basic data of recommendation modeling and can be directly obtained from a knowledge-intensive service platform. An employer-service matrix, as shown in fig. 3, represents an employer-service relationship, with 1 in the matrix indicating that there is an interaction between the employer and the service (e.g., the employer has purchased the service too much), and 0 indicating that there is no interaction between the employer and the service. The task of the application is to find the service most interesting to the employer among the services not interacted with the employer and recommend to the employer.
2) Worker-service relationship: the worker-service relationships may also be obtained directly from the knowledge-intensive service platform, as shown in FIG. 3, which shows what type of knowledge-intensive service each worker may provide.
3) Service-service knowledge similarity relationship: the knowledge-intensive service platform typically classifies the knowledge-intensive services provided on the platform according to specific rules and sorts out service-category relationships. Although different platforms may have different classification rules, services or sub-categories under the same category often have some similarity between them. Knowledge similarity between services can be quantified based on service-category relationships. The service-category relationships in FIG. 3 show a typical service-category relationship tree, with the lowest S representing an independent knowledge-intensive service, C 2 Representing category II, C 1 Representing a first class category. According to the characteristics of the platform service category relation, S, C in the service-category relation can be used 2 Or C 1 Knowledge data is collected for keywords to quantify knowledge similarity between services. Such as:
the information describing the knowledge needed to complete the knowledge intensive service is typically text information. In C 2 Knowledge data is collected for keywords and the data collected under the same keywords are combined into one text.
Using Paraggraphvector (Le)&Tomas mikolov, 2014) model learning each C 2 Vector representation of knowledge text below, resulting in
Calculation ofThe inner product of the two to obtain C 2 Knowledge similarity between them.
According to C in service-class 2 S relation, C 2 Mapping knowledge similarity between servicesService-service knowledge similarity is obtained.
4) Business activity-service relationship: s, C in service-category relation based on value chain consultation model 2 Or C 1 Categorizing under different businesses and establishing business-to-service relationships. Such as:
c, according to a value chain consultation model 2 Categorizing into different business activities, establishing A-C 2 Relationship.
According to C in service-class 2 S relation, A-C 2 The relationship maps to the service to obtain a business-service relationship.
And S13, linking all object relations to construct a heterogeneous information network of the knowledge intensive service.
And linking all objects according to the network template of the knowledge intensive service information network established in the step S11 and the object relationships of employer-service, worker-service, service-service knowledge similarity, business activity-service and the like established in the step S12, and establishing high-order connectivity among heterogeneous objects. A segment of a knowledge-intensive service heterogeneous information network is shown in fig. 4. In the network, the underlying employer-service interactions are represented by thick solid lines, and other worker-service, service-service knowledge similarities, business-service, and like object relationships are represented by thin solid lines. Recommendations generated by exploring multiple relationships between objects in heterogeneous information networks are represented by thick solid lines. In building heterogeneous information networks for knowledge-intensive services, thresholds for varying service-service knowledge similarity control the richness of interactions in the network. And searching for knowledge similarity with best model performance through experiments.
In step S1, the method for obtaining the heterogeneous polynary relationship in the heterogeneous information network specifically includes: and designing a meta-path, and acquiring heterogeneous multi-element relations in the heterogeneous information network according to the designed meta-path by adopting a random walk strategy. The semantic interpretation of each meta-path is shown in table 1 and the structure of each meta-path is shown in fig. 5.
Table 1 element Path semantic interpretation
Given element pathAlong the meta-path->The defined relationships may obtain many specific paths in the heterogeneous information network of the knowledge-intensive service, which are the path instances, denoted p. The set of path instances linking employer u and service s becomes a meta-path based heterogeneous multi-relationship between u and s. In heterogeneous information networks of knowledge-intensive services, exploring heterogeneous multi-relationships among objects along meta-paths may generate recommendations. As in the heterogeneous information network shown in fig. 3, based on path instance u 1 -s 1 -u 2 -s 2 ,u 1 -s 1 -u 3 -s 2, and (u1 -s 1 -a 1 -s 2 ) sim Likely to be to employer u 1 Recommended service s 2 . Based on path instance u 4 -s 3 -w 2 -s 4 and (u4 -s 3 -a 2 -s 4 ) sim Likely to be to employer u 4 Recommended service s 4 。
If the heterogeneous information network of knowledge intensive services is not built, only depending on employer-service bipartite graph, employer u 4 Will not be recommended for any service. The rich heterogeneous multi-relations contained in the heterogeneous information network of knowledge-intensive services are more advantageous to reveal relationships between employer knowledge-intensive services and to grow recommendations than bipartite graphs containing only employer-service interactions.
S2, merging the heterogeneous polynomials, and learning comprehensive vector representations of employers and services;
in step S2, the method for learning the comprehensive vector representation of the employer and the service by fusing the heterogeneous multi-relations includes the following steps:
s21, merging heterogeneous multi-element relations acquired by each element path, and learning employer and service semantic specific vector representations;
in step S21, the specific method for learning the employer and the semantic-specific vector representation of the service is to fuse the heterogeneous polynary relations acquired by each element path: inputting the heterogeneous polynary relation obtained by each element path into a meta 2vec model for learning to obtain employer and service semantic specific vector representation, wherein if N element paths are designed, the N element paths are represented as follows:after meta 2vec model learning, each node in the heterogeneous information network may obtain N sets of semantic-specific vector representations, extracting the vector representations of employers and services, expressed as: />And
s22, merging the heterogeneous multi-element relations from all element paths, and learning the comprehensive vector representation of employer and service.
In step S22, the method for learning the employer vector representation includes the following steps:
s221: the employer semantic specific vector representation is converted by a multi-layer perceptron, specifically:
s222: calculating the importance of each meta-path to each employer based on the employer semantic-specific vector representation: given semantic attention vector x and employer u in-element pathEmployer semantic specific vector on +.>With x and->Is to represent meta-path +.>Importance to employer u, express +.>
Where W, b and x are the weight matrix, bias term and semantic attention vector, respectively, h is the parameter to be learned and is shared among all meta-paths and semantic specific node vectors.
S223: normalizing the importance of each element path to the employer through a softMax function, and calculating the weight of each element path: meta pathThe weight for employer u is denoted +.>
The higher the meta-path->The more important it is to employer u. Typically each meta-path is differently weighted for different employers;
s224: weighted fusion based on the weight of each element pathGet a comprehensive vector representation of employers, useAs weight coefficients, employer synthetic vector representations are specifically:
note that, the meta pathThe weight for all employers is denoted +.>Can be expressed as all->Average value:
it should be noted that, the process of learning the service synthesis vector representation is the same as the process of learning the employer synthesis vector representation by fusing heterogeneous polynomials.
In step S3, the predicted value is obtained through a recommendation model based on the comprehensive vector representation of the employer and the service, where the recommendation model specifically includes:
wherein ,representing predicted value, p u Representing employer complex vector representations, q s Representing a service synthesis vector representation, h being the prediction layer.
It is to be noted that,before obtaining the predicted value through the recommended model, optimizing and training the model by using a loss function, and minimizing the predicted valueAnd the actual value R us Difference between them. The application uses a non-sampling strategy optimization model proposed by (EfficientNeurMatrixFactorizationWithoutsamplinorRecommendation) and the like, and the model loss function is as follows:
where B ε U represents a group of employers in an employer set, S is a service set, d is the dimension of the embedded vector, c us R represents us WhereinRepresenting the weight of the negative sample.
And S3, acquiring a predicted value based on the comprehensive vector representation of the employer and the service, and recommending the service to each employer according to the predicted value.
In step S3, the method for recommending services to each employer according to the predicted value includes: and sorting according to the size of the predicted value to obtain a recommendation list, and recommending the first K services of the recommendation list to the employer.
In summary, the knowledge intensive service recommendation method based on heterogeneous multivariate relation fusion provided by the application can well realize employer knowledge intensive service requirement mining in the industrial field. A large number of specialized applications are gathered on the knowledge-intensive service platform, and the knowledge-intensive service recommendation method is provided for the recommendation system, so that non-professional users facing information overload can find knowledge-intensive services interested by themselves in the large number of specialized applications, the viscosity of employers to the platform is increased, sales are promoted, and the service quality and profit of the platform are improved.
For a better understanding of the solution of the present application, the following will further illustrate the present application by taking the service recommendation of the eight-pig platform as an example.
The eight-ring-in-pig network (zbj.com) is the leading platform in china and the greatest focus on service-knowledge-intensive service transactions. Currently, more than 2 tens of millions of users are aggregated on a platform. And various knowledge-intensive service services related in the production and management processes of enterprises such as industrial intellectual construction, engineering design, industrial and commercial tax, brand design, market research, game development, software development and the like are provided. The proposed method was tested on knowledge-intensive service transaction data of the eight-ring platform to evaluate the above method, with the specific objective of answering the following research questions:
problem one: how does the performance of the proposed algorithm model of the present application?
And a second problem: when learning the vector representation of employer and service in fusion of heterogeneous multi-relations, which meta-paths are relied upon by employer nodes and service nodes, respectively?
Problem three: how does the model performance change with changes in the relationship under consideration?
Fourth problem: how does model performance change with knowledge similarity?
Problem five: in a non-sampling strategy, coefficientsHow does model performance be affected?
Data set:
transaction data: employer-service transaction records 16219 were obtained from the pig ring network, with 2355 employers, 1875 services, each of which purchased at least 5 services, each of which was purchased at least 3 times. The sparsity of the data was 0.37%. Employer-service interaction records are sparse and inadequate.
Service document: as shown in fig. 3, the service document obtained from the pig ring network contains a service-category relationship and a worker-service relationship. In the service category relation, services under the same secondary category basically need similar knowledge to complete, but knowledge similarity still exists among part of secondary categories. The worker-service relationship shows which services each worker can provide.
Recruitment data: because of the knowledge similarity between the secondary categories, 7900 recruitment data is collected from the recruitment site as knowledge data with 200 secondary category names in the service-category relationship as keywords. Wherein each keyword has at least 30 pieces of related knowledge data, and each piece of knowledge data comprises two contents of post responsibilities and post requirements. The job responsibilities describe primarily the work content and the job requirements describe primarily the knowledge and capabilities required.
All data were randomly divided into 3 groups, 80% as training set, 10% as validation set, and 10% as test set.
Experiment setting: when the random walk collection path example is based on the meta-path, the walk times w of each node are 200, and the walk length l is 20. The magnitude k of the heterogeneous domain is 7 when learning the semantic specific vector representation of each node using meta 2vec fusion of the polynomials from each meta-path. In the whole learning process, the input and output dimensions of the vector are 128. In the experimental process, analyzing the coefficients in the service-service knowledge similarity sim and non-sampling strategyAnd the effect of meta-path combination variations on model performance.
Evaluation index: two indices widely used for recommendation system evaluation were used: the recovery@K and the NDCG@K. The recovery@K represents the probability that the positive case in the sample is correctly predicted. Ndcg@k is an index for evaluating the accuracy of ranking of a recommendation table, and the more the recommendation list matches the preference of a user, the higher the value of ndcg@k. To prevent randomness and volatility, an average of 10 experimental index values was selected as the final test value of the index.
The comparison scheme is as follows:
BPR (BayesianPersonalizedRanking): a personalized ordering model based on Bayesian posterior optimization can minimize the loss of pairwise ordering.
NeuMF: a typical recommendation algorithm based on deep learning. The matrix decomposition is combined with the multi-layer perceptron, so that the model has nonlinear expression capability, and complex user-object interaction can be modeled better.
NGCF (NeuralGraphCollaborativeFiltering): a recommendation model based on a graph neural network. Co-embedded propagation, co-signals in the user-object bipartite graph are explicitly encoded in the form of high-order connectivity.
LightGCN-linearly propagates user and item embeddings in a user-item interaction graph and represents the final embeddings of the user and item as weights for the embeddings learned by the layers.
EASE: a model without a hidden layer has good performance in the aspect of processing sparse data.
Slimmelastic: by solving the problem of 1 and l2 The regularization optimization problem learns the sparse aggregate sparse matrix to produce high quality recommendations.
Experimental results:
the method comprises the following steps: when the meta-path combination is [ USAS ] sim ，USUS，USWS]Knowledge similarity is 0.45, coefficient in non-sampling strategyAt a value of 0.02, the proposed recommendation system has the best recommendation performance. As shown in tables 2 and 3, the recovery@k of the proposed model was improved by 24.46% -47.56% and the ndcg@k index was improved by 46.67% -58.7% compared to the best performing model in the baseline model.
TABLE 2 values of model recovery@K index
TABLE 3 values of model NDCG@K index
Influence of meta-path combination:
when the model performance is best, the meta-path combination is [ USAS ] sim ，USUS，USWS]. Meta-path groupThe combined employer and service weights are [0.0026,0.7768,0.2205 ]]And [0.9674,0.0215,0.0110 ]]. Representation learning dependency path USAS for specifying service nodes sim This verifies the correctness of the relationship definitions 1.2 and 2.2. The representation of employer nodes learns the dependency paths USUS and USWS, which verifies the correctness of the relationship definition 3.1. In the employer node representation learning process, although the meta-path USAS sim But we analyze that the service node passes through the meta-path USAS sim Semantics are aggregated and then the service nodes affect semantic aggregation of employer nodes via meta-path USUS.
Ablation experiment:
this experiment compares the impact of model performance presented by different meta-path combinations. Fig. 6 shows the performance of the combined model at different meta-paths. As can be seen from fig. 6, as the number of relationships to be considered increases, the more useful information is obtained in model training, the better the performance of the model. However, too many relationships are considered, too much noise is also generated, which affects further improvement of the model performance. The composite constraint between the relationships is considered to effectively filter noise in the relationships, thereby helping to further improve model performance.
Influence of knowledge similarity on model performance:
fig. 7 shows the model performance at different similarities. As can be seen from fig. 7, the performance of the model increases with the similarity of knowledge, and decreases after the best performance is achieved. The knowledge similarity value that best performed the model is approximately 0.45.
Coefficients in non-sampling strategyInfluence on model performance:
in non-sampling strategiesThe coefficients control the negative (missing value) weight in the samples. As can be seen from FIG. 8, the model performance is as follows +.>The increase of (1) rises first, reaches a maximum and then starts to fall, when>The model performs best when the value of (2) is around 0.02. This confirms the study by (c.chemet., 2020) et al that for coefficient data, will +.>Setting to a smaller value is beneficial to alleviating the unbalanced learning problem and improving the model performance.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present application, and not for limiting the same; although the application has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some or all of the technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit of the application, and are intended to be included within the scope of the appended claims and description.
Claims (5)
1. The knowledge intensive service recommendation method based on heterogeneous multi-element relation fusion is characterized by comprising the following steps of:
s1, constructing a heterogeneous information network and acquiring a heterogeneous multi-element relation in the heterogeneous information network;
s2, merging the heterogeneous polynomials, and learning comprehensive vector representations of employers and services;
s3, acquiring predicted values based on comprehensive vector representations of employers and services, and recommending services to each employer according to the predicted values;
in step S1, the method for constructing the heterogeneous information network includes the following steps:
s11: constructing a network mode of the knowledge intensive service information network to illustrate object types and interaction relations thereof in the knowledge intensive service information network; wherein the object types include employers, services, knowledge, business, and workers, and the interaction relationships include employer-services, knowledge-services, business-services, and workers-services;
s12, acquiring original data, preprocessing the original data, and establishing an object relation; wherein the object relationships include employer-service relationships, business-service relationships, worker-service relationships, service-service knowledge similarity relationships;
s13, linking all object relations based on a network mode of the knowledge intensive service information network to construct a heterogeneous information network of the knowledge intensive service;
in step S2, the method of learning a comprehensive vector representation of employer and service, incorporating the heterogeneous multi-relations, comprises the steps of:
s21, merging heterogeneous multi-element relations acquired by each element path, and learning employer and service semantic specific vector representations;
s22, merging heterogeneous multi-element relations from all element paths, and learning comprehensive vector representations of employers and services;
in step S3, the method for obtaining the predicted value based on the comprehensive vector representation of the employer and the service is as follows: constructing a recommendation model based on the comprehensive vector representation of employers and services, and acquiring a predicted value based on the recommendation model, wherein the recommendation model is specifically:
wherein ,representing predicted value, p u Representing employer complex vector representations, q s Representing service comprehensive vector representation, h being a prediction layer;
in step S3, the method for recommending services to each employer according to the predicted value is as follows: and sorting according to the size of the predicted value to obtain a recommendation list, and recommending the first K services of the recommendation list to the employer.
2. The knowledge intensive service recommendation method based on heterogeneous multi-relation fusion according to claim 1, wherein in step S1, the method for obtaining heterogeneous multi-relation in heterogeneous information network specifically comprises: and designing a meta-path, and acquiring heterogeneous multi-element relations in the heterogeneous information network according to the designed meta-path by adopting a random walk strategy.
3. The knowledge intensive service recommendation method based on heterogeneous multi-relation fusion according to claim 1, wherein in step S21, the specific method for learning employer and service semantic specific vector representations by fusing heterogeneous multi-relations acquired by each meta-path is as follows: and inputting the heterogeneous polynomials acquired by each element path into a meta 2vec model for learning to obtain employer and service semantic specific vector representations.
4. The knowledge intensive service recommendation method based on heterogeneous multi-relationship fusion according to claim 1, wherein the method of learning the complex vector representation of employers in step S22 comprises the steps of:
s221: converting the employer semantic-specific vector representation by the multi-layer perceptron;
s222: calculating the importance of each meta-path to each employer based on the employer semantic-specific vector representation;
s223: normalizing the importance of each element path to each employer, and calculating the weight of each path;
s224: and carrying out weighted fusion based on the weight of each element path to obtain the comprehensive vector representation of the employer.
5. The knowledge intensive service recommendation method based on heterogeneous multivariate relation fusion according to claim 1, further comprising optimizing training a model by using a loss function to minimize a predicted value before obtaining the predicted value by recommending the modelAnd actual valueR us Difference between them.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111276405.4A CN113837847B (en) | 2021-10-29 | 2021-10-29 | Knowledge intensive service recommendation method based on heterogeneous multivariate relation fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111276405.4A CN113837847B (en) | 2021-10-29 | 2021-10-29 | Knowledge intensive service recommendation method based on heterogeneous multivariate relation fusion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113837847A CN113837847A (en) | 2021-12-24 |
| CN113837847B true CN113837847B (en) | 2023-08-29 |
Family
ID=78966679
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111276405.4A Active CN113837847B (en) | 2021-10-29 | 2021-10-29 | Knowledge intensive service recommendation method based on heterogeneous multivariate relation fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113837847B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115146963B (en) * | 2022-07-01 | 2023-06-30 | 西南交通大学 | Multi-link linkage coordination method based on third-party cloud platform |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2017041541A1 (en) * | 2015-09-08 | 2017-03-16 | 北京邮电大学 | Method for pushing recommendation information, and server and storage medium |
| CN110598130A (en) * | 2019-09-30 | 2019-12-20 | 重庆邮电大学 | Movie recommendation method integrating heterogeneous information network and deep learning |
| CN112182424A (en) * | 2020-11-11 | 2021-01-05 | 重庆邮电大学 | Social recommendation method based on integration of heterogeneous information and isomorphic information networks |
| CN112800342A (en) * | 2021-04-15 | 2021-05-14 | 中国人民解放军国防科技大学 | Recommendation method, system, computer device and storage medium based on heterogeneous information |
-
2021
- 2021-10-29 CN CN202111276405.4A patent/CN113837847B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2017041541A1 (en) * | 2015-09-08 | 2017-03-16 | 北京邮电大学 | Method for pushing recommendation information, and server and storage medium |
| CN110598130A (en) * | 2019-09-30 | 2019-12-20 | 重庆邮电大学 | Movie recommendation method integrating heterogeneous information network and deep learning |
| CN112182424A (en) * | 2020-11-11 | 2021-01-05 | 重庆邮电大学 | Social recommendation method based on integration of heterogeneous information and isomorphic information networks |
| CN112800342A (en) * | 2021-04-15 | 2021-05-14 | 中国人民解放军国防科技大学 | Recommendation method, system, computer device and storage medium based on heterogeneous information |
Non-Patent Citations (1)
| Title |
|---|
| Heterogeneous Graph Embedding for Cross-Domain Recommendation Through Adversarial Learning;Jin Li等;《international conference on database systems for advanced applications DASFAA2020:database systems for advanced applications》;第507-522页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113837847A (en) | 2021-12-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| AU2023206202A1 (en) | Risk identification and risk register generation system and engine | |
| US6836773B2 (en) | Enterprise web mining system and method | |
| US7117208B2 (en) | Enterprise web mining system and method | |
| CN110245285B (en) | Personalized recommendation method based on heterogeneous information network | |
| Nethravathi et al. | Business intelligence appraisal based on customer behaviour profile by using hobby based opinion mining in India: a case study | |
| WO2011018245A1 (en) | Method for scoring items using one or more ontologies | |
| JP2013536484A (en) | Computerized Agent System and User Instruction Semantic Networking | |
| van Capelleveen et al. | The influence of knowledge in the design of a recommender system to facilitate industrial symbiosis markets | |
| Lukosius et al. | Marketing theory and big data | |
| CN114861050A (en) | Feature fusion recommendation method and system based on neural network | |
| CN115221387A (en) | Enterprise information integration method based on deep neural network | |
| CN113837847B (en) | Knowledge intensive service recommendation method based on heterogeneous multivariate relation fusion | |
| US11144938B2 (en) | Method and system for predictive modeling of consumer profiles | |
| Darena et al. | Machine learning-based analysis of the association between online texts and stock price movements | |
| Paul et al. | A weighted hybrid recommendation approach for user’s contentment using natural language processing | |
| CN116362620A (en) | Electronic commerce online customer service evaluation method | |
| Lee et al. | How to reduce confirmation bias using linked open data knowledge repository | |
| Tang et al. | Service recommendation based on dynamic user portrait: an integrated approach | |
| Agarwal et al. | Binarized spiking neural networks optimized with Nomadic People Optimization-based sentiment analysis for social product recommendation | |
| Ahmed et al. | Amazon Product Recommendation System | |
| Ramesh et al. | Improvement to Recommendation system using Hybrid techniques | |
| Kumar et al. | Identifying meaningful neighbors for an improved recommender system | |
| Arunagiri et al. | Selection of browsers for smartphones: a fuzzy hybrid approach and machine learning technique | |
| Liao et al. | Mining the co-movement in the Taiwan stock funds market | |
| Snášel et al. | User profiles modeling in information retrieval systems |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |